Friends, work is getting so hilarious right now! My agents run in a loop with /goal. When they need context, they DM one of my colleagues on Teams. But some of my colleagues are also using agents, so their agent replies to mine. My agent is basically asking another person’s agent for context that is not in either person’s brain. It’s somewhere in their emails, documents, meetings, or cloud storage. The other agent finds it, sends it back, and mine keeps working toward my goal. It feels obvious that agent-to-agent communication is about to become insanely fast. Most of it is just context exchange. The next step is a company brain where all the context lives, with access rights attached. Instead of messaging five people, your agent retrieves exactly what it is allowed to know and continues working. The funniest part: I was grabbing a coffee when a colleague received a Teams message “from me.” He looked at me, saw that I was obviously not typing, and immediately understood that it was my agent. We are already becoming managers of agents talking to other agents.

Muse is winning at model-market fit. I haven’t seen anything it does today that ChatGPT Work couldn’t do months ago, or that Instinct or another agent couldn’t do. But using it feels different. It’s fast, smooth, and you actually want to keep using it. A faster, cheaper model paired with a great harness makes a huge difference. For an agent taking dozens of steps to get something done, every bit of latency adds up. You feel it every time you ask for something and if it's too slow it will feel clunky and annoying. And cost matters to the product experience too. It determines how much work you can afford to run in the background, how often the agent can check things, and how many attempts it can make. It's so obvious that they have a cheap model because they run the Feed updates hourly, they also create memories hourly too and not nightly like other agents. You can have the same capabilities on paper and end up with very different products. IMO that’s what I think Muse is getting right: the underlying model is very good.

Very funny going out in SF now: one guy heads home in a Zoox, another swears by Waymo, and I’m taking a Tesla Robotaxi. Nobody even mentioned Uber or Lyft. Autonomous rides are so smooth and comfortable. The future feels amazing.

It’s funny, Meta went from having my Instagram and WhatsApp data to now having access to my email, calendar, DoorDash, Amazon and pretty much everything. In the last 24 hours, it bought me socks, ordered my Whole Foods groceries, booked a cleaning service and got me a burger for dinner. Meta’s last disclosed North American Facebook ARPU was around $227/year, largely from ads. I suspect it can push that number significantly higher now that it understands not only what I look at, but what I need, what I buy and what I’m planning to do. Also the much bigger opportunity might be becoming the aggregation layer between me and the entire internet. If Meta can take even a tiny percentage of the commerce it facilitates, or of the money it saves me, this could become enormous! It already saved me $200 by canceling subscriptions and services I no longer needed. This feels much bigger than better ad targeting. Ads are useful but giving me money back is better imo. One additional thought: the agent is increasingly making the decisions for me. I knew nothing about that burger place. The agent researched it, told me which burger I should order, and I just said “okay” without giving it much more thought. Agents are becoming the decision makers in both B2C and B2B. Increasingly, every business will be selling not just to humans, but to their agents. Everything becomes B2A: business to agents.

The Powerball choice is actually a pretty fun finance problem: $450.5M today vs $1.04B paid over 29 years. If you take the $450M and put it in the S&P 500, the break even return is only around 5% a year. At 8%, it becomes roughly $4B after 29 years. But there’s also just a massive preference for the present. $450M today is infinitely more useful than money I may receive 20 or 30 years from now. I can invest it, give it away, build things with it, whatever. And, slightly morbid point, I might not even be alive in 29 years (who knows?!) Give me the $450M today and an S&P 500 ETF and I will manage 😂

Kids growing up today are so lucky. With chatbots, they basically have an infinitely patient teacher who will follow their curiosity anywhere. I personally spend hours at night going down random history rabbit holes, asking dumb questions, or having science explained to me until it finally clicks. Imagine having that at 10 years old.

If you were in San Francisco during the crypto token boom, the AI token boom feels weirdly familiar, especially at the extremes. Extreme displays of wealth. This constant paranoia that you need to make generational money right now or somehow you’ll end up poor. Everyone around you seems to be getting richer, so even if you’re doing objectively well, you start feeling like you’re the only idiot missing the trade haha Then come the shady deals, insiders making absurd money, people optimizing every relationship for upside, very little loyalty, and a surprising amount of betrayal the second incentives change. It creates this strange atmosphere where ambition, FOMO, greed and genuine technological excitement are all mixed together. The huge difference is that AI is actually an extraordinary underlying technology. The economic value being created is real, the productivity gains are real, and the benefit to humanity is enormous!! So from my experience, the boom can feel unhealthy at the edges while the thing we’re collectively building is incredibly healthy and important.

I think most software will be created by AI, and increasingly most software will also be consumed by AI on our behalf. So we’re going to see an explosion of software, not a reduction. Looking at my own usage, I estimate that: • For every web search I do, AI does 1,000x more • For every email I send, AI sends 100x more • For every SMS I send, AI sends 10x more • For every spreadsheet I open, AI probably queries/manipulates 1000x more data behind the scenes The conclusion is pretty simple: we should start building software for agents, because agents are going to become the primary users.

Buying a Matic vacuum cleaner was my best purchase last year. It’s so good. Now, it has voice command functionality, which is the best because voice is the future computer interface. Once you get a Matic, it’s easy to envision a future where your house is filled with robots that clean and cook for you. It’s coming.

Ultra-fast LLM latency is cool because it unlocks real-time apps where speed is make-or-break like SRE (diagnose outages while they’re still unfolding), e-commerce (buyers hate latency), gaming (fluid NPC dialogue that doesn’t break immersion) etc! Kudos to @cerebras!

Ok friends, sharing the recipe to build basically any agent product: • Memory: learn everything useful about the user, their work, people, preferences, projects, decisions, etc. from their emails, drive, computer files etc • Context engineering: figure out which tiny subset of all that memory + current state actually belongs in the context window for the task. • Skills: teach the agent how to do the work. Build a complex DCF. Prepare a board meeting. Triage an inbox. Run customer research. Write an investment memo. • Tools / MCPs: give it access to the world. Email, calendar, Slack, GitHub, CRM, browser, databases, internal APIs, computer use, etc. • Agent runtime: the loop that plans, acts, observes, retries, delegates, checkpoints state and can keep working for hours or days. • Triggers: cron jobs + events. New email arrives. Meeting ends. Customer churns. Metric changes. Deadline approaches. The agent wakes up without being prompted. • Trust / permissions: know what it can do autonomously, what requires approval, and whose identity / permissions it is acting with. • Evals: trace everything and continuously measure whether the agent is actually getting better. Then optimize the hell out of memory quality, context selection, action quality, latency and token cost. That’s basically the agent stack. If you need more verticalization (aka finance, logistics etc. just add more MCPs and skills).

At what point do we admit that all AI agent products are kinda similar: connect your email, calendar, docs, Slack, cloud, whatever. The agent will then extracts a ton of context about how you work, who you work with, how you communicate, what you care about to turn that into markdown memory files & then an orchestrator on top that learns your patterns and tries to reproduce them. The frontier isn’t really “chat” anymore. It’s proactive background agents. Users don’t prompt the AI. Agent notices something needs to happen and does the work on the behalf of the user. And underneath, a lot of it is surprisingly simple conceptually: give the model more context, give it tools, memory, and some cron jobs / triggers / routines. So the actual game is becoming: who can build the best memory, take the most useful actions, and do it while burning the fewest tokens. Because it’s actually pretty easy to build an amazing personal memory system if you’re willing to spend $100+ per user. The hard part is getting 95% of that quality for $5. The goal is the highest quality context and actions, lowest inference cost to get the highest gross margin.

It's not from Anthropic but from the European Union. The official text is Regulation (EU) 2024/1689 which is a sort of GDPR for AI. "2. Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated. Providers shall ensure their technical solutions are effective, interoperable, robust and reliable as far as this is technically feasible, taking into account the specificities and limitations of various types of content, the costs of implementation and the generally acknowledged state of the art, as may be reflected in relevant technical standards. This obligation shall not apply to the extent the AI systems perform an assistive function for standard editing or do not substantially alter the input data provided by the deployer or the semantics thereof, or where authorised by law to detect, prevent, investigate or prosecute criminal offences."

100% epic. Prometheus rising at Starbase feels like the emergence of a new American aesthetic: classical, monumental, technologically optimistic and entirely unafraid of ambition